my-project-public

repository

loading code, commits, and activity

repositories

loading repo index

#1	# Prompt Engineering Guide
#2
#3	## Introduction
#4
#5	Effective prompts are the foundation of powerful AI agents. This guide teaches you how to craft system prompts that produce consistent, high-quality results.
#6
#7	## Prompt Structure
#8
#9	### Basic Framework
#10
#11	```
#12	[IDENTITY] + [CAPABILITIES] + [GUIDELINES] + [FORMAT] + [EXAMPLES]
#13	```
#14
#15	### Identity
#16
#17	Define who/what the agent is:
#18
#19	```
#20	You are a DeFi security auditor specializing in Ethereum smart contracts.
#21	```
#22
#23	Clear identity creates consistent behavior and sets user expectations.
#24
#25	### Capabilities
#26
#27	List specific skills and knowledge:
#28
#29	```
#30	CAPABILITIES:
#31	- Analyze Solidity code for common vulnerabilities
#32	- Check for reentrancy, overflow, and access control issues
#33	- Review token economics and minting mechanisms
#34	- Assess gas optimization opportunities
#35	```
#36
#37	### Guidelines
#38
#39	Behavior rules and constraints:
#40
#41	```
#42	GUIDELINES:
#43	- Classify risks as Critical/High/Medium/Low
#44	- Provide code snippets showing vulnerabilities
#45	- Suggest specific fixes with examples
#46	- Never guarantee 100% security
#47	- Recommend professional audits for production code
#48	```
#49
#50	### Output Format
#51
#52	Structure for responses:
#53
#54	```
#55	FORMAT:
#56	Security Score: [0-100]
#57	Critical Issues: [List]
#58	Recommendations: [Prioritized steps]
#59	Code Examples: [Where applicable]
#60	```
#61
#62	### Examples
#63
#64	Demonstrate desired behavior (optional but powerful):
#65
#66	```
#67	EXAMPLE:
#68	User: "Check this function for issues"
#69	You: "Security Score: 65/100
#70
#71	Critical Issues:
#72	1. Reentrancy vulnerability on line 45
#73	- Current: function withdraw() external { ... }
#74	- Fix: Add nonReentrant modifier
#75
#76	Recommendations:
#77	1. Implement checks-effects-interactions pattern
#78	2. Add reentrancy guard
#79	3. Consider using OpenZeppelin's ReentrancyGuard"
#80	```
#81
#82	## Writing Clear Prompts
#83
#84	### Be Specific
#85
#86	❌ Vague: "You help with DeFi"
#87	✅ Clear: "You analyze yield farming strategies across Ethereum Layer 2 protocols, focusing on risk-adjusted returns"
#88
#89	### Use Action Verbs
#90
#91	✅ Analyze, Calculate, Compare, Evaluate, Recommend, Monitor, Optimize
#92
#93	### Set Boundaries
#94
#95	```
#96	IN SCOPE:
#97	- Ethereum mainnet and L2s (Arbitrum, Optimism, Base)
#98	- Established protocols (TVL > $10M)
#99	- Yield strategies up to medium risk
#100
#101	OUT OF SCOPE:
#102	- Unaudited contracts
#103	- New protocols (< 3 months old)
#104	- High-risk leveraged positions
#105	- Tax or legal advice
#106	```
#107
#108	## Advanced Techniques
#109
#110	### Conditional Logic
#111
#112	```
#113	IF user provides wallet address:
#114	- Fetch on-chain data
#115	- Analyze current positions
#116	- Provide personalized recommendations
#117
#118	IF user asks about specific protocol:
#119	- Check TVL and audit status
#120	- Analyze recent activity
#121	- Provide risk assessment
#122
#123	IF query is unclear:
#124	- Ask 2-3 clarifying questions
#125	- Don't make assumptions
#126	```
#127
#128	### Tone and Personality
#129
#130	```
#131	COMMUNICATION STYLE:
#132	- Professional but approachable
#133	- Use analogies for complex concepts
#134	- Acknowledge uncertainty when appropriate
#135	- Encourage user questions
#136	```
#137
#138	For DeFi agents, balance between:
#139
#140	- Technical accuracy (for experienced users)
#141	- Accessibility (for newcomers)
#142
#143	### Error Handling
#144
#145	```
#146	WHEN USER REQUEST IS OUT OF SCOPE:
#147	"I specialize in yield optimization for established protocols. For [user's topic], I recommend consulting [appropriate resource]."
#148
#149	WHEN DATA IS UNAVAILABLE:
#150	"I don't have real-time data on that protocol. I can explain the general mechanism, or you can check [reliable source]."
#151
#152	WHEN RISKS ARE HIGH:
#153	"⚠️ High Risk Alert: This strategy involves [specific risks]. Only proceed if you understand and can afford potential losses."
#154	```
#155
#156	## Domain-Specific Tips
#157
#158	### DeFi Prompts
#159
#160	Include:
#161
#162	- Risk disclaimers
#163	- Gas cost considerations
#164	- Smart contract audit status
#165	- Impermanent loss warnings
#166	- Slippage and MEV awareness
#167
#168	```
#169	RISK DISCLAIMER:
#170	Always append to financial recommendations:
#171	"Not financial advice. Smart contract risks exist. Only invest what you can afford to lose. DYOR."
#172	```
#173
#174	### Trading Agents
#175
#176	```
#177	NEVER:
#178	- Guarantee profits or outcomes
#179	- Provide specific entry/exit prices
#180	- Recommend leverage without risk warnings
#181	- Ignore market context
#182
#183	ALWAYS:
#184	- Include multiple scenarios (bull/bear/sideways)
#185	- Show supporting data/charts
#186	- Explain reasoning transparently
#187	- Suggest position sizing
#188	```
#189
#190	### Technical Agents
#191
#192	```
#193	CODE REVIEW STRUCTURE:
#194	1. High-level assessment
#195	2. Line-by-line analysis
#196	3. Security concerns
#197	4. Gas optimizations
#198	5. Best practice recommendations
#199	6. Suggested improvements with code
#200	```
#201
#202	## Testing Your Prompts
#203
#204	### Test Suite
#205
#206	Try these scenarios:
#207
#208	1. Happy Path: Standard, clear request
#209	2. Ambiguous: Vague or unclear query
#210	3. Out of Scope: Request beyond agent's expertise
#211	4. Edge Case: Unusual or complex situation
#212	5. Adversarial: Attempt to break agent behavior
#213
#214	### Iterative Refinement
#215
#216	```
#217	Version 1: Basic prompt
#218	↓
#219	Test with users
#220	↓
#221	Identify issues: Too verbose? Missing context? Inconsistent?
#222	↓
#223	Version 2: Refined prompt
#224	↓
#225	Test again
#226	↓
#227	Version 3: Optimized prompt
#228	```
#229
#230	## Common Pitfalls
#231
#232	### ❌ Too Broad
#233
#234	"You help with cryptocurrency"
#235	→ Too vague, leads to inconsistent behavior
#236
#237	### ❌ Too Narrow
#238
#239	"You only calculate APY for Uniswap V2 ETH/USDC pools on Thursdays"
#240	→ Overly specific, not useful
#241
#242	### ❌ Contradictory
#243
#244	"Be concise. Provide detailed explanations with examples."
#245	→ Conflicting instructions confuse the model
#246
#247	### ❌ Assuming Knowledge
#248
#249	Referring to concepts without defining them
#250	→ Explain or link to definitions
#251
#252	## Prompt Templates
#253
#254	### General Purpose
#255
#256	```
#257	You are [ROLE] with expertise in [DOMAIN].
#258
#259	Your core functions:
#260	1. [Function 1]
#261	2. [Function 2]
#262	3. [Function 3]
#263
#264	When responding:
#265	- [Guideline 1]
#266	- [Guideline 2]
#267	- [Guideline 3]
#268
#269	Format your responses as:
#270	[Structure description]
#271	```
#272
#273	### Analytical Agent
#274
#275	```
#276	You are a [TYPE] analyst for [DOMAIN].
#277
#278	ANALYSIS FRAMEWORK:
#279	1. Data Collection: [Sources]
#280	2. Processing: [Methods]
#281	3. Interpretation: [Approach]
#282	4. Recommendation: [Format]
#283
#284	DELIVERABLES:
#285	- Summary (2-3 sentences)
#286	- Detailed analysis
#287	- Risk assessment
#288	- Actionable insights
#289
#290	CONFIDENCE LEVELS:
#291	- High: Strong supporting evidence
#292	- Medium: Reasonable evidence
#293	- Low: Limited or conflicting data
#294	```
#295
#296	### Educational Agent
#297
#298	```
#299	You are a [TOPIC] educator specializing in [AUDIENCE LEVEL].
#300
#301	TEACHING APPROACH:
#302	- Start with fundamentals
#303	- Build complexity gradually
#304	- Use analogies and examples
#305	- Check understanding with questions
#306	- Provide resources for deeper learning
#307
#308	FORMAT:
#309	1. Concept explanation (ELI5)
#310	2. Technical details
#311	3. Practical example
#312	4. Common misconceptions
#313	5. Further reading
#314	```
#315
#316	## Resources
#317
#318	- [OpenAI Prompt Engineering Guide](https://platform.openai.com/docs/guides/prompt-engineering)
#319	- [Anthropic Prompting Guide](https://docs.anthropic.com/claude/docs/prompt-engineering)
#320	- [Learn Prompting](https://learnprompting.org/)
#321
#322	## Examples from Top Agents
#323
#324	See `/src/` directory for 505 example agents with proven prompts across various domains.
#325
#326	---
#327
#328	Questions? Open an issue or join our Discord community.
#329
#330
#331

z6Mkq5mY3JWtxoxUobWcfNHm7AkRubgSWEZTkBVqZXJviFZ5/my-project-public